The system will be going down for maintenance on Wednesday 22nd March 7-9pm NZT. Apologies for the inconvenience.
Developing a high definition video quality of experience model based on influential parameters : a thesis presented in partial fulfilment of the requirements for the degree of Doctor of Philosophy in Computer Science at Massey University, Palmerston North, New Zealand
Multimedia applications are quickly becoming a basic necessity for both business and
personal situations. They are being used on daily basis, for a big part of the population,
it's becoming the main source of entertainment and communication. Internet based
Television and IPTV are now being preferred over standard Television due to the added
functionality that it provides. Users not only expect good quality they also require it to
be available continuously without fail. The research reported in this thesis is focused
on the Development of a high definition video Quality of Experience (QoE) model
based on influential parameters". We have proposed a model for predicting QoE for
high definition videos. We sought and justified appropriate parameters for inclusion into
our prediction model. Our analysis has helped us in identifying parameter quantization
as the most influential parameter from content domain. Whereas the parameters such
as packet loss, packet reorder and jitter were found to be equally influential on QoE
from the network domain. We also were able to show that parameter buffer size was
the most influential parameter from the customer premises processing domain. We
also showed that by integrating these parameters from the three domains of content,
network and customer premises processing has enabled us to predict QoE with greater
accuracy. The model enables the service provider to predict QoE and act before user
perception goes beyond a predetermined threshold level. By including all the most
influential parameters from all three domains we have ensured that we were accounting
for a complete end to end effect of a user's perception. We were also able to identify
the optimal configuration for good quality. In addition, we determined configurations
that reduced the output quality to unacceptable levels that should be avoided. We
were also able to find out a mathematical relationships between these parameters.
We have discussed the current state-of-the-art knowledge for QoE in the three identified domains and then proceeded to identify the shortcomings of models and methods
found in the current literature. We made use of Taguchi robust designs for reducing the
run time of experiments. This also helped us in analyzing many different configurations
but only conducting experiments on a subset of these configurations. For three experiments
the Taguchi DoE technique helped us reduce the number of test combinations
from 81 to only 9 combinations. For the fourth experiment we were able to reduce the
number of combinations from 256 to only 18 combinations. We were able to reduce
known boredom and memory effects by careful design of our experiments. Moreover,
we were able to analyze the results using an appropriate signal to noise ratio (SNR)
analysis method. This method helped us identify the most influential parameters for
QoE in the three identified domains and it also assisted in the identification of optimal
configurations. The method utilize quality loss model for identifying optimum configurations.
By this method we found the SNR for each configuration and those individual
parameter configurations were selected which ensured the highest SNR values. This
helped identify the optimum configurations from each domain. In addition, we used
the House of Quality (HoQ) method to validate the results of our SNR analysis. We
were able to show that for individual domain the HoQ method was able to verify the
results of our SNR analysis. HoQ only failed when we tried to use it to verify the end to
end video quality degradation. We identified that the reason for HoQ method's failure
in generating acceptable results was the use of weights to identify the most influential
parameters in situation were all parameters under investigation where already found
to be influential. These heavy weight parameters canceled the effect of each other and
the parameters became statistically insignificant.
We proposed a QoE prediction model by using the most influential parameters
identified from each domain. Our results show that these influential parameters were
the major reasons for degradation in quality. Hence, using such parameters can help
in better prediction of video quality degradation.
We made use of ordinal logistic regression for developing the prediction model. We
proposed a complete model which was able to predict quality upto an 88% accuracy.
Later, we developed a reduced version of the model for low computing solutions which
was able to predict quality with 84.5% accuracy.